Datasets, Code & Challenges
Datasets and benchmarks I have created or co-created, and the challenges built around them. Please check each dataset’s licence and cite the associated paper if you use it.
Datasets
LTDv2 — Long-Term Thermal Drift
A large-scale thermal surveillance dataset for studying multi-object detection under long-term concept drift: seasonal change, weather, sensor re-calibration, and time-of-day shift across an extended capture period. Built to test whether detectors hold up when reality moves away from the training distribution.
Download (Hugging Face) · Paper (TechRxiv) · Code & starting kit
Parola, Aakerberg, Johansen, Nikolov, Cimino, Nasrollahi & Moeslund (2025).
SVSR — Spatially Variant Super-Resolution benchmark
1,119 real low-resolution images degraded by complex noise of varying intensity and type, each paired with real noise-free ×2 and ×4 high-resolution ground truth. Built to test how robust real-world super-resolution methods are when degradation varies across the image rather than being uniform; it also works as a denoiser benchmark.
Download (Zenodo, 2.1 GB) · Paper (WACV 2024) · DOI: 10.5281/zenodo.10044260 · Licence: CC BY-NC-SA 4.0
Aakerberg (2023). Companion benchmark to PDA-RWSR.
RELLISUR — Real Low-Light Image Super-Resolution
12,750 paired images: real low-light low-resolution captures with normal-light high-resolution references, across multiple resolution scales and degrees of underexposure. Closes the gap between low-light enhancement and super-resolution, which the literature had treated separately even though real-world visibility is usually limited by both at once.
Published at NeurIPS 2021, Datasets & Benchmarks Track. Now over 2,000 downloads.
Download (Zenodo, 8.9 GB) · Paper (NeurIPS 2021) · DOI: 10.5281/zenodo.5234969 · Licence: CC BY 4.0
Aakerberg, Nasrollahi & Moeslund (2021).
Challenges
RTIOD — Robust Thermal-Image Object Detection Challenge
Hosted at the 6th Real-World Surveillance workshop, WACV 2026. Participants build detectors that stay accurate as thermal imagery drifts over months and seasons, evaluated on held-out periods from LTDv2. A YOLO starting kit and COCO-format dataset loader are provided.
Challenge (Codabench) · Repository & starting kit · Challenge report
Real-World Surveillance Workshop & Challenge — IEEE/CVF WACV
I co-organise the Real-World Surveillance workshop, now in its seventh edition, along with its associated challenge. The workshop brings together researchers working on surveillance problems as they actually occur: uncontrolled capture, degraded imagery, long deployment horizons, and operational constraints.